<p>Cost overrun is one of the most prevalent challenges in construction projects, significantly affecting investment efficiency and project management performance. As construction projects become increasingly large-scale and complex, the development of accurate predictive models plays a crucial role in supporting cost risk management. This study proposes a machine learning based approach for predicting construction project cost overruns using the Extreme Gradient Boosting (XGBoost) algorithm integrated with the SHAP (SHapley Additive exPlanations) interpretability technique. The dataset consists of 1000 simulated observations, incorporating key characteristics of construction projects, including project size, estimated project cost, material costs, schedule pressure, delay risk, design changes, and macroeconomic indicators such as the construction material price index and inflation rate. The model was optimized using Randomized Search combined with 5-Fold cross-validation to enhance predictive performance. Model accuracy was evaluated using Mean Absolute Error (MAE) and the coefficient of determination (R²). The results demonstrate that the XGBoost model achieves high predictive accuracy and effectively identifies the most influential factors contributing to cost overruns. Furthermore, SHAP-based analysis enhances model interpretability, providing valuable insights to support decision-making in construction cost management.</p>